UNIVERSITÀ POLITECNICA DELLE MARCHE ________________________________________________________________________________________________________________________________________________________

DIPARTIMENTO DI SCIENZE ECONOMICHE E SOCIALI

POLITICAL INSTITUTIONS BEHIND

GOOD GOVERNANCE

DAVID BARTOLINI RAFFAELLA SANTOLINI

QUADERNO DI RICERCA n. 405

ISSN: 2279-9575

Settembre 2014

Comitato scientifico:

Renato Balducci Marco Gallegati Alberto Niccoli Alberto Zazzaro Collana curata da: Massimo Tamberi

Abstract

The present work looks at the role of political institutions — political regimes and elect-oral rules — in determining the performance of the government to define and implementsound policies for the economy. The results of the empirical investigation on a panelof 80 democracies over the period 1996-2011, show an important impact of the politicalregime on the performance of the government — the presidential regimes reduces thequality of the government —, while electoral rules do not matter. However, the analysisshows that the interaction between political regimes and electoral rules plays a crucialrole for the quality of the government. In particular, a presidential regime improvesthe government performance when associated with a majoritarian rule, while worsens itwhen combined with a proportional rule.

JEL Class.: D72, H11Keywords: electoral rule; political system; government effectiveness;

regulatory quality

Political institutions behind good governance∗

David Bartolini Raffaella Santolini

1 Introduction

The capacity to the government to implement programs and policies is crucial for theeconomic development of countries and the well-being of citizens; it is the governmentthat issues regulations for the market to work in an efficient way, and provides most ofthe public goods and infrastructures. Too much discretion, however, can lead to rent-seeking behaviour on the part of government officials. When devising the institutionalsetting within which the executive operates two distinct and sometimes conflicting needsemerge: on the one hand, the need to restrain the executive in order to avoid “abuses”and rent-seeking behaviour; on the other hand, political institutions should guaranteethe discretion and freedom to implement policies and economic programs. The differencein the performance of the executive among democratic countries, can be attributed to adifferent mix between delegation of powers and accountability.

The trade-off between restrain and delegation of powers is well understood in Aghionet al. (2004) who considers the share of votes that can block the action of the governmentas a measure of discretion. The authors maintain that the “optimal” level of restraindepends on several characteristics of the system, such as the degree of polarisation insociety, the aggregate benefit of the policy, the individual degree of risk aversion, theavailability and efficiency of fiscal transfers, the degree of protection of property rights.Also the political science literature is well aware of the problem, and the analysis ofpolitical institutions (presidential vs parliamentary regimes) is conducted on the basisof the political features that affect the trade-off between implementation and restrain ofgovernment actions (Samuels and Shugart, 2003). For instance, Mainwaring and Shugart(1997) consider the different features of the political system and their impact on the well-functioning of the presidential system, showing that the presidential system works betterwhen presidents have weak legislative powers, the assembly is not highly fragmented,and elected politicians follow party discipline.

In the present work, we focus on two institutional features that affect the mix betweendelegation of powers and accountability: the electoral rule and the political regime (par-liamentary vs. presidential). The former affects the way in which the assembly is formed— party fragmentation, selection of candidates, etc. —, while the latter affects the in-

∗Ackowledgements: We are grateful for useful comments from an anonymous referee. We would liketo thank Eniel Ninka for his assistance in data collection. The views expressed are those of the authorsand do not reflect those of the OECD or its Member countries.

1

centives of the executive and its relationship with the assembly (Mainwaring, 1993). Wemaintain that it is from the interaction of the political regime with the electoral rulethat the actual powers and restrains of the government emerge. The interaction is cru-cial for the performance of the government. For instance, one of the main criticisms ofthe presidential regime is the risk of immobilism due to a president not supported bythe assembly; this risk is lower in a parliamentary regime where a vote of no confid-ence would quickly remove the executive. Immobilism, however, is not only induced bythe political regime, it mainly depends on the composition of the assembly. A majorit-arian rule tends to generate a two-party system that makes it easier for the president toachieve stable and large majorities. By contrast, the combination of presidential regimeand proportional electoral rule makes it more difficult to build stable coalitions, thushindering the implementation of government policies and jeopardising the performanceof the government (Mainwaring, 1993).

The aim is to empirically investigate the combined effect of the political institutionson the government ability to formulate and implement sound policies efficiently, sheddinglight on the most appropriate electoral rule to associate with a political regime. Figure1 offers some preliminary indications in this regards. The figure compares governmentsability to devise and implement sound policies for economic development across demo-cratic countries, making use of two perception-based indicators developed by the WoldBank within the World Governance Indicator project (Kaufmann et al., 2010). The firstindicator, government effectiveness, captures the perception about the quality of publicservices and polices and the credibility of the government’s commitment to such policies.The second indicator, regulatory quality, captures the perception of the government toprovide policies that foster private sector development. Together they provide an indic-ation of the quality of the government both in terms of its ability to devise good policiesand to implement them. The average performance of the countries in our sample showsa positive correlation between quality and the combination of a parliamentary regimewith a proportional electoral rule,1 as well as a positive correlation between quality andthe combination of a presidential regime with a majoritarian electoral rule.

Empirical analysis is based on a panel of 80 democracies over the period 1996-2011.The results of the estimation of the dynamic specification of the econometric model, showthat electoral rules do not affect the quality of the government, while the political regimehas a statistically significant impact on quality, that is, the governments in presidentialregimes display lower performance than governments in parliamentary regime. But,most importantly, we find robust evidence that the role of electoral rules is crucialfor the performance of political regimes. Indeed, the estimation results show that theperformance of the government improves when the presidential regime is associated witha majoritarian electoral rule, whereas worsens it when combined with a proportional rule.

Similar to our work Panizza (2001), using a cross-section, finds a negative correlationbetween government effectiveness and the presidential regime, and no significant impact

1As conjectured by Lijphart, “the combination of parliamentarism with proportional representationshould be an especially attractive one to newly democratic and democratizing countries”(Lijphart, 1991,p. 72).

2

Figure 1: The average impact of political institutions on government performance

Note: The values of the columns refer to the average over time (1996-2011) and over countries of theindexes of Government Efficiency and Regulatory quality, respectively. Both indicators range between-2.5 and 2.5, with higher values indicating stronger governance performance. The column labelledpresidential-majoritarian refers to the average of the sub-sample of countries with a presidential regimeand a majoritarian electoral rule; the same interpretation is valid for the rest of the columns.

of the electoral rule. Our analysis, while confirming his results using panel data anda dynamic specification, shows that the electoral rules are actually important for theperformance of the political regime. The importance of this interaction is found also byKunicova and Rose-Ackerman (2005) which study the relationship between the politicalsystem and the level of corruption, using a cross-section of ninety-four democracies. Theyshow the presidential regime associated with a proportional electoral rule determines highlevels of corruption. The latter finding complements our results about the importanceof matching the political regime with the electoral rule.

The rest of the paper is organised as follows: in the next section we investigatethe possible explanations for the effects produced by political institutions on the gov-ernments’ ability to initiate and sustain policy reforms; then, in Sections 3 and 4, wedescribe the dataset and the methodology used to conduct our empirical investigation,respectively; in Section 5 we present the results of our empirical analysis and check fortheir robustness; finally, concluding remarks are provided in Section 6.

2 The political framework

In this section, we consider the two institutional features that are at the basis of ouranalysis: the electoral rule and the political regime.

The electoral rule determines the way in which politicians are selected and affectsthe composition of the assembly. Although the details of the electoral rules differ amongcountries, it is customary to classify electoral rules into two broad categories: major-

3

itarian and proportional. Majoritarian rules are characterised by a large number ofsingle-member districts where the candidate with the largest percentage of votes is se-lected (“the first past the post” rule). A proportional rule is generally characterised bya small number of voting districts (e.g. multi-member districts) or a single nationwidedistrict (as in the Netherlands) where all members of the legislature are elected by aproportional representation rule. In this case, the seats in the assembly are distributedaccording to the share of votes received by each candidate. Thus the voting districtmagnitude (i.e., the number of candidates elected in the voting district) under the ma-joritarian rule is one, while under the proportional rule is larger than one. A differentcombination of district magnitude produces a mixed electoral rule.

One way in which the electoral rule can affect the performance of the government isthrough the composition of the assembly. A majoritarian rule tends to produce a less(party) fragmented assembly than a proportional rule. The different degree of party frag-mentation affects government stability and its ability to implement policies. The major-itarian electoral rule makes a two-party system more likely to emerge from the electoralcompetition (Duverger, 1954; Shugart and Carey, 1992). Bipartitism favours stabilityof government and its decision-making ability. Governments supported by a large andhomogeneous majority have high probability to remain in power for the entire durationof electoral mandate, during which are able to implement their programs (Lardeyret,1991). By contrast, proportional rules produce fragmented assemblies, which leads toexecutives supported by a wide coalition of parties; this situation is likely to result infrequent elections, and thus short-lived governments (Duverger, 1954; Lardeyret, 1991;Persson et al., 2004). The proportional rule, however, provides better political repres-entation of minorities. Lijphart (1999) argues that a proportional (or consensual) modelof democracy, based on proportional representation with multi-member districts, shouldbe preferred to a majoritarian democracy when societies are heterogeneous in terms ofethnolinguistic, ideological, religious, and racial aspects.2 The proportional rule allowsminorities to actively participate in political debates and affect the decision making pro-cess, providing an effective representation of their interests. Moreover, the empiricalinvestigation conducted by Lijphart (1999) shows that the consensual democracy is alsoassociated to a better government performance in terms of lower inflation and unemploy-ment rate, better control of violence, and higher quality of the democracy. The trade-offbetween effectiveness and representation can be summarised as: “proportionalists tendto attach greater importance to the representativeness of government, while pluralityadvocates view the capacity to govern as the more vital consideration”(Lijphart, 1991,p.76).

The electoral rule can play a role also in the restrain and discipline of the electedpolitician. The recent literature on political agency maintains that a majoritarian ruleis more likely to reduce rent-seeking behaviour than a proportional rule, because of

2Examples of consensual democracy are Belgium and Switzerland, while examples of majoritariandemocracies are Australia, Canada, the United Kingdom, and New Zealand. The latter form of govern-ment is commonly referred to as the Westminster model, consisting of plurality rules, bipartitism, andone-party governments, among other features.

4

Table 1: Majoritarian vs proportional representation rule

Majoritarian Proportional

Government-type single-party party coalitionOpposition-type cohesive fragmentedRepresentation low highGovernment stability high lowRent-seeking attitudes low high

stronger party competition in key (marginal) districts (Persson and Tabellini, 1999).When the electoral rule is taken in isolation, with no consideration for the institu-

tional feature of the political system, it is not possible to determine its ultimate impacton political and economic outcomes. It is important, however, to determine the differen-tiated impact of each electoral rule on the specific characteristics of the political system.As summarised in Table 1, the majoritarian rule tend to foster government stabilitybecause of lower party fragmentation, while producing a strong opposition and low rep-resentation of minority groups. The proportional rule is associated with a fragmentedassembly, a weak opposition, and high representation of minority groups.

The impact of the electoral rules on the performance of the government depends alsoon the relationship between the assembly and the executive power. This link is definedby the political regime, presidential or parliamentarian, which shapes the relationshipbetween the executive and the assembly, as well as the normative structure within whichthe executive operates. In a parliamentary regime, the power of the executive originatesfrom the majority in the legislative assembly, therefore its survival depends on the voteof confidence of the assembly. This makes members of the assembly more disciplined insupporting the executive, but not enough to prevent frequent changes of government.In a presidential system, both the assembly and the president are directly elected, andthe assembly cannot influence the composition or the life of the government, likewise thepresident cannot remove member of the parliament. The presidential regime is character-ised by greater separation between the assembly and the executive than a parliamentaryregime. Mainwaring and Shugart (1997) points out that the members of congress in apresidential regime can act in the assembly without worrying about the consequences oftheir choices on the life of the executive. By contrast, in many parliamentary regimes,voting on specific legislative matters is often associated with a vote of confidence forthe executive, thus members of the parliament that would oppose the specific legislativeact, but want the government to survive, are forced to vote in favour. Furthermore, theindependence of assembly and executive may also increase accountability as respons-ibilities are more clearly associated with the executive or parties in the assembly thatsustain the president, than in the parliamentary system. At the same time, independ-ence could generate immobilism, because the loss of confidence of the assembly towardsthe executive cannot be easily resolved with a change of government.

5

Mainwaring and Shugart (1997) provides and excellent discussion of the role of polit-ical regimes in determining the performance of the government. The authors contrastthe presidential and the parliamentary system showing that there are several charac-teristics of the political system that affect their efficacy. In particular, they maintainthat the presidential regime performs better when the president has weak legislativepowers, parties are disciplined, and the assembly is not fragmented in many politicalparties. In an influential article, Linz (1990) considers several potential drawbacks ofthe presidential system, among which is the claim that the presidential system intro-duces rigidity in the system, as the president cannot be removed until the end of themandate. While it is true that the presidential system raises the cost for removing theexecutive, it is also true that the instability associated with frequent changes of govern-ment jeopardise the possibility of the executive to implement policies. There is again atrade-off between restraining an executive that does not implement the desired policiesand allowing the executive the time and means to implement such policies. In the samearticle, Linz maintains that the presidential system introduces a winner-take-all type ofpolitics. This is, however, not only the result of the political regimes (i.e., the directelection of the president), but also the impact to the electoral rule on the assembly. Aspreviously mentioned, the majoritarian electoral rule tends to produce a two-party sys-tem where the party which gets more votes wins the representation of the district. Againit appears that it is the interaction between the political regime and the electoral rulethat defines the characteristics of political institutions and ultimately the efficacy andefficiency of the governmental action. Mainwaring and Shugart (1997) maintain thatlegislative independence is particularly problematic when associated with high partyfragmentation. An electoral rule that produces less party fragmentation (such as themajoritarian voting rule) may reduce the problem of a deadlock between the presidentand the parliament. In this way, a presidential system associated with a majoritarianrule can avoid immobilism.

From the analysis of the economic and political science literature it emerges theimportance of both electoral rules and political regime for the performance of the gov-ernment. It is the interaction between these two features that determines the quality ofthe government. In particular, we can formulate the following hypothesis to be testedin the empirical part of the paper:

- the presidential system induces a better (worse) government performance whenassociated with a majoritarian (proportional) rule.

3 Data and variables

The empirical analysis is conducted on a sample of 80 democratic countries over theperiod 1996-2011. Democratic countries are selected according to the political rightsand civil liberties ranking provided by the Freedom in the World. We selected countrieswith a score between 1 and 5. This selection rule is then restricted by including inthe sample only countries that are classified as electoral democracies according to the

6

Freedom in the World Survey.3

The government ability to implement and formulate policies is measured by the indic-ators related to government effectiveness (GovEff) and regulatory quality (RegQual),developed by the World Bank (Kaufmann et al., 2010).4 As explicitly indicated inKaufmann et al. (2010), the two indicators cover the aspect of governance related to“the capacity of the government to effectively format and implement sound policies”. Indetail, GovEff reflects citizens’ perception over the quality of public services, bureau-cracy and the degree of civil service independence from political pressure. The indicatorRegQual reflects citizens’ perception over the quality of rules and regulation imple-mented to foster the economic performance of the private sector. The quality of policyimplementation is better captured by GovEff , while the quality of policy formulation isbetter represented by RegQual, as maintained by Panizza (2001). The two indexes rangebetween -2.5 to 2.5, with the highest value indicating a strong governance performance.The World Governance Indicator project gathers information and data from a wide vari-ety of sources, including surveys of firms and households and subjective assessments ofthe business community, public sector and non-governmental organisation.

The political institution indicators are based on the Database of Political Institutions2012 (DPI hereafter) released by The World Bank (Beck et al., 2001). The nature ofthe political regime is captured by the dummy variable PRES which assumes value oneif the form of government is presidential and zero if it is parliamentarian. Countriesare coded as presidential if the president is elected either directly by citizens or by anelectoral college, and there is no prime minister. Countries with both a president anda prime minister are classified as presidential if the president has a veto power to blocklegislation, and the assembly needs a supermajority to overcome the veto. Furthermore,a political regime is coded as presidential if the president holds both the power to appointand dismiss the prime minister and/or the other executive members, and the power todissolve the assembly and call for new elections. According to this typology, our sampleconsists of 32 countries with a presidential regime and 48 countries with a parliamentaryregime.

Electoral rules are classified as proportional, majoritarian or mixed. The latter caseemerges when features of the majority and proportional rules are combined to elect gov-ernment representatives. The analysis is conducted with two dummy variables. Thefirst dummy, PR, assumes value one if candidates are elected on a proportional basisand zero otherwise; the second dummy variable, MAJ, assumes value one if candidatesare elected using the majoritarian rule and zero otherwise. Along with these two dummyvariables we use the mean district magnitude (MDM), which captures the whole spec-trum of electoral rules in a continuous way. The MDM indicator consists in the weighedaverage of the number of representatives elected in each electoral district. It assumesvalue between 1 and infinity. A value of 1 corresponds to “pure” majoritarian elect-

3See Freedom House (2009), Electoral Democracies: Freedom in the World 1989-90 to 2009. Fordetails on data and methodology see http://www.freedomhouse.org/report-types/freedom-world andhttp://www.freedomhouse.org/report/freedom-world-2012/methodology.

4Missing data for 1997, 1999 and 2001 are replaced with data computed as inter-temporal average ofthe year immediately before and after.

7

oral rule, while larger values of MDM denote increasing proportional electoral rules.In our sample three countries have an average district magnitude lower than 1, namelyAustralia, Botswana and Namibia. According to the DPI methodology, when the in-formation on the number of representatives per district is not available, it is replaced bythe number of parliamentary seats dived by the number of districts. Therefore, if thereare more districts than seats it must be that less than 1 person is elected on average inall electoral districts.

In order to test our main hypothesis, the indicator of the political regime (MAJ)is interacted with the two dummies representing the two electoral rules considered, andwith the district magnitude indicators. Accordingly, MAJPRES assumes value 1 forcountries with a presidential regime associated to a majoritarian electoral rule; PRPRESassumes value 1 for countries with a presidential regime associated to a proportionalelectoral rule; and MDMPRES represents the interaction of presidential countries withthe mean district magnitude indicator.

The empirical analysis is complemented with the inclusion of other variables thatcould affect the performance of the government. An important institutional featureis the degree of decentralisation of the public administration.5 The expected impact ofdecentralisation on government performance is not well defined. On the one side it shouldbe positive as local politicians can provide policies that are closer to citizens’ needs, andare in general more accountable than “distant” central government politicians. On theother side, policy spillovers require more co-ordination at the local level, overlapping ofcompetences may hinder the performance of the government, and local politicians canbe an easier target for special interest groups (Treisman, 2002). In order to accountfor this phenomenon an indicator of regional autonomy, provided in the DPI database,is included in the empirical specification. Another important determinant of quality,linked to the decentralisation, is the risk that local politicians may pursue goals that arein contrast with the central government (i.e., vertical competition), and this may reducethe overall performance of the government. Enikopolov and Zhuravskaya (2007) showthat the presence of “strong” national parties reduces this risk. Therefore, we includein our estimation strategy the average of the age of the two largest government partiesand the main opposition party as a proxy for party strength, we label this variableParty age. We control also for the maturity of a democracy using the variable Tensys,obtained from the DPI database. This variable accounts for the consolidation of thedemocratic institution, controlling for the fact that mature democracies are characterisedby institution more likely to be stable, and more experienced policy makers than newlyformed democracies. We label this variable Demo.

As seen in the previous section, an important dimension for the performance ofpolitical institution is the degree of homogeneity of society and the need to cater tominorities. The basic ideas is that less heterogenous societies will suffer less from thisproblem. We include in the analysis the degree of population homogeneity by consideringthe percentage of the largest religious and ethnic group in the population, using data

5See Kyriacou and Roca-Sagales (2011), Bartolini and Santolini (2013).

8

released by the CIA World Fact Book.6

Other control variables are the logarithm of the population size, Pop, and the growthof population living in urban areas, Urban. Both indicators should have a negativeimpact on government performance because the larger the population the higher theprobability of having ethno-linguistic, religious and income differences, while the pro-cess of urbanisation introduces growing challenges in terms of metropolitan governance(Bartolini, 2014).

The economy does have an impact on the quality of the government, thus we in-troduce the logarithm of the per-capita Gross Domestic Product (GDP) evaluated atconstant 2000 US Dollars, and the degree of trade openness (Openness) measured asthe sum of exports and imports of goods and services as a share of GDP. Both GDPand trade openness should impact positively the perceived performance of the govern-ment (de Mello and Barenstein, 2001). An additional control for impact of the economicsituation of a country, we consider the consumer price index (CPI). The motivation isthat government may be pushed by concerns of rising prices to adopt swift measure tocontrast the rising inflation and thus recover efficiency in the economic system with anoverall improvement of the quality of governance.

Some of the control variables may be affected by a problem of endogeneity. In partic-ular, although the quality of the institutions may depend on the economic developmentof a country, it is also possible that better political institutions lead to growth in GDP,as pointed out by the extensive literature on the economics of institutions. The sameis true for trade openness. A reverse causality problem can also concern political partystrength and democratic consolidation, as pointed out in Enikopolov and Zhuravskaya(2007). Therefore, we instrument these variables using their temporally lagged values.The descriptive statistics of the dependent and the control variables are illustrated inTable 2, while a detailed description of each variable is presented in the Appendix.

4 The dynamic panel data model

In this section we illustrate the empirical model and methodology used to estimate theeffects produced by electoral rules and political regimes on the performance of the gov-ernment. The analysis is conducted separately for the two indicators of governmentperformance. We exploit the panel dimension of our sample estimating the autoregress-ive fixed-effects model illustrated in equation (1), where the dependent variable Qi,t

corresponds to the two indicators of government performance, GovEff and RegQual,for country i in period t. The autoregressive model includes the first-order lagged valueof the dependent variable Qi,t−1 among the regressors and a 1× k vector of explanatoryvariables x′it = (x1it, ..., x

kit). The dynamic specification seems to be more appropriate

because the government performance exhibits a great deal of persistence. The vari-ables MAJ and PRES correspond to the majoritarian rule and presidential system,respectively. Fixed effects fi are included to control for the omission of unobserved

6Data on ethnic group homogeneity are missing for 12 countries, namely, Benin, Denmark, France,Italy, Madagascar, Malawi, Malta, Mozambique, Papua New Guinea, Portugal, Spain, and Sweden.

9

Table 2: Descriptive statistics

Variable Obs Mean Std. Dev. Min Max

GovEff 1280 0.49 0.94 -1.21 2.34RegQual 1280 0.52 0.82 -1.59 2.08PR 1280 0.47 0.50 0 1MAJ 1280 0.24 0.43 0 1MDM 1271 14.65 37.27 0.72 450PRPRES 1278 0.19 0.39 0 1MAJPRES 1278 0.07 0.26 0 1MDMPRES 1269 0.60 1.09 -0.33 6.11PRES 1278 0.39 0.49 0 1Party age 1269 39.98 33.97 1 191Pop 1280 3.640E+07 1.290E+08 1.007E+05 1.200E+09Urban Pop growth 1280 1.47 1.45 -8.42 5.14Religion homog. 1186 71.23 22.67 13.8 99.8Ethnic homog. 1015 76.98 17.99 28 99.4GDP pc 1262 10325.97 11880.05 142.39 56285.30Openness 1260 86.48 42.28 14.93 333.53CPI 1242 96.14 27.79 3.34 210.91Autonomy 1280 0.19 0.39 0 1Demo 1277 28.03 25.16 1 81Low income 1280 0.26 0.44 0 1

characteristics of countries, while time effects τt capture undefined shocks common tocountries. Finally, a constant term, a, is included in the model along with an error termε which is normally distributed with zero mean and constant variance.

Qit = a+ ρQit−1+αMAJit + δPRESit + βx′it + fi + τt + εit (1)

(i = 1 . . . N ; t = 1 . . . T )

Since our main interest is the effects produced by the interaction between the electoralrule and the political regime, we include the interaction term MAJPRES among theregressors of the dynamic panel model (equation 2).

Qit = a+ ρQit−1 + αMAJit+φMAJPRESit + δPRESit + βx′it + fi + τt + εit (2)

(i = 1 . . . N ; t = 1 . . . T )

The econometric model is based on the majoritarian dummy which indicated theimpact with respect to the other electoral rules (i.e., mixed and proportional). As arobustness check we also conduct the analysis substituting the majoritarian dummywith the proportional rule dummy PR and the district magnitude variable MDM.

Both empirical specifications are estimated using the System Generalised Method ofMoments (Sys-GMM) estimator (Arellano and Bover, 1995; Blundell and Bond, 1998)since it performs well for large N and small T . Moreover, it accounts for problems of

10

endogenous variables that can be present in our panel. The Sys-GMM estimator is basedon a system of equations in levels and in differences, with instruments in first differencesfor equations in levels, and instruments in levels for equations in first differences.7 TheSys-GMM estimator is an extended version of the first-differenced (Diff-) GMM estimatorproposed by Arellano and Bond (1991). However, it is more efficient and has smaller biasin finite sample than the first differencing GMM estimator8 In spite of these advantages,Bun and Windmeijer (2010) have found that the equation in levels may suffer froma weak instrument problem when the variance ratio is large and series are persistent,revealing the existence of a potential weak instrument problem also for the Sys-GMMestimator.

Besides this problem, the estimator may suffer from proliferation of instruments(Roodman, 2009) which may arise when the p-values of the Hansen test and the Difference-in-Hansen test are closed to 1 (Roodman, 2009). The Hansen test checks the validityof the full set of instrumental variables, while the Difference-in-Hansen test checks thevalidity of the sub-set of instruments used in the level equations. Although both tests areconsistent in the presence of heteroscedasticity and autocorrelation in the error terms,they are weaker when many instrumental variables are used in a panel regressions. Inthis circumstance, although the validity of the instruments is confirmed, “the potentialfor false-positive results is serious” (Roodman, 2009, p.156). In order to solve this prob-lem “results should be aggressively tested for sensitivity to reductions in the numberof instruments”(Roodman, 2009, p.156). The number of instrumental variables can besignificantly reduced by collapsing instruments and limiting lag length. However, it isdifficult to select the optimal number of instruments. A thumb rule generally adopted inempirical works is Z ≤ N , that is, the number of instruments Z needs to be less than orequal to the number of cross sections N (Roodman, 2009). Moreover, Bun and Wind-meijer (2010) have found that the equation in levels may suffer from a weak instrumentproblem when the variance ratio is large and series are persistent, revealing the existenceof a potential weak instrument problem also for the Sys-GMM estimator.

Finally, the consistency of the Sys-GMM estimator depends on the second-orderserial correlation in the first-differenced residuals. In other words, it is consistent whenthe second-order autocorrelation is not detected in the first-differenced residuals. Thiscondition can be tested using the statistics m2, developed by Arellano and Bond (1991).9

5 Results

The results of the estimation of the econometric models, illustrated in Tables 3-6, showthat the dynamic specification is more appropriate than the static one, since the coeffi-

7The lagged values of both the predetermined and strictly exogenous variables can be additionalinstruments.

8A drawback of the Diff-GMM estimator is that it suffers from a problem of weak instruments whenthe coefficient of the lagged dependent variable is near to unity and/or the variance ratio of the fixedeffects to the idiosyncratic errors increases (Blundell and Bond, 1998; Bun and Windmeijer, 2010).

9Also the test statistic m1 developed by Arellano and Bond (1991) is used to test the absence of thefirst-order autocorrelation in the first-differenced residuals which implies serially correlated disturbances.

11

cient of Qt−1 is always statistically significant. The set of instrumental variables used isvalidated by the Hansen J test that accepts the null-hypothesis with few exceptions (seecolumn 2 of Table 5 and column 3 of Table 6). The Difference-in-Hansen test accepts thenull of the validity of the subset of instruments. The Sys-GMM estimates are consistentsince the m2 test always accepts the null-hypothesis in the presence of second-orderautocorrelation in the first-differenced residuals. According to the F-test reported ineach table, all estimated coefficients are jointly statistically significant.

Tables 3 and 4 present the estimation of model (1), without the interaction term.The results show that electoral rules do not significantly affect the performance of thegovernment with respect to the political regime. By contrast, the presidential regime re-duces both indicators of the government effectiveness and regulatory quality. This resultmay be the consequence of immobilism due to the presence of strong checks and balance.If the president does not hold the majority of seats in the assembly, the process requiresa direct confrontation with all parties in the assembly in order to gather a consensusaround each policy the executive wants to implement. The political stalemate betweenthe assembly and president can defer the initiation of any policy reforms, worseninggovernment performance overall.

The estimation of model (2) with interaction term confirms the importance of con-sidering, at the same time, the political regime and the electoral rule. The analysisestablish an important role for the electoral rule as determinant of the impact of thepolitical regime on the quality of the government.

The empirical analysis shows that the adoption of a majoritarian electoral rule ina presidential regime significantly improves government effectiveness. In Table 5 thecoefficient of the interaction term MAJPRES describes the difference in the effect ofadopting a majoritarian rule versus a proportional or mixed rule in a presidential sys-tem. The coefficient is positive and statistically significant at 10% level, although onlyfor GovEff . This result is corroborated by the negative and significant coefficient as-sociated to the interaction with the proportional electoral rule, PRPRES, in column2 of 5. Similarly, as shown in Table 6, the coefficients of PRPRES and MDMPRESare negative and significant. These results are consistent with the political frameworkdiscussed in Section 2. A more fragmented assembly makes it more difficult to form andmaintain a stable majority in support of the executive. In this setting a majoritarianrule reduces party fragmentation allowing the president to control and influence the le-gislation and executive activity through the members of his party for a time sufficientto implement his programme.

Furthermore, the estimated coefficients of the interaction terms are robust becausethe F-test shows that the coefficients α, φ and δ are jointly different form zero.

As regards the control variables, the estimation of model (1) shows that GDP percapita is positively and significantly correlated with GovEff . Trade openness has alsoa positive but modest impact on both indicators of performance. These results confirmthe hypothesis that more open and economically developed countries display better gov-ernment performance. The CPI indicator is positively and significantly correlated withRegQual but its impact is very small. As expected, urban population growth reduces

12

government performance because of increasing complexity. The coefficients associatedwith the degree of homogeneity of a country have a mild impact on the indicator of gov-ernment efficiency, while no statistically significant impact on regulatory quality. Alsothe indicator associated with regional autonomy is rarely significant. The only signi-ficant result is the positive impact on regulatory quality, implying that more regionalautonomy may improve the quality of policy formulation. This result maybe explainedby the higher level of political accountability associated with autonomous regions, wherepolicymakers set regulations in order to create a sound business environment for attract-ing firms and workers (Weingast, 1995; Qian and Weingast, 1997). Fianlly, the variableassociated with democratic tenure does not have a significant impact on the performanceof the government.

The significant impact of the control variables remains substantially unchangedwith the introduction of the interaction terms, with the exception of the coefficientsof Openness and Demo that are no longer statistically significant, and Partyage thatis positively and significantly correlated with the regulatory quality.

5.1 Robustness analysis

The lower performance of the presidential regime with respect to the parliamentaryregime could be imputed to the composition of the sample. As noted by Panizza (2001),among the most developed countries only the United States is a presidential regime, thusthe possibility that this result is driven by the overrepresentation among less developedcountries of the presidential regime.

The World Bank classifies countries in four categories according to their income: low,mid-low, upper-middle, and high income. According to this classification our sampleconsist of 15 presidential countries classified as low and mid-low and 17 presidentialcountries classified as upper-middle and high income countries. The distribution in oursample suggest that our data do not suffer from a sample bias. In order to better check forthis bias, we included in the empirical analysis a dummy variable that assumes value 1 ifthe country is classified by the World Bank as a low or lower-middle income country, andzero otherwise. The coefficient of the dummy is never statistically significant, reducingthe risk that our results are driven by a sample bias.

In order to conduct further robustness checks, we estimate both econometric equa-tions with a sub-sample of countries from which countries with an average GDP percapita below 500 US$ (about the 10th lowest percentile) are excluded.10

Estimating model (1) with the subsample of 72 countries11, shows that the negativeeffect of the presidential regime on GovEff disappears, whereas the negative impact onRegQual, detected in Table 4, remains statistically significant. As regards the electoral

10Eight countries are dropped, namely, Bangladesh, Benin, Ghana, Madagascar, Malawi, Mali, Mol-dova and Mozambique. A part from Bangladesh and Moldova, the other countries have a presidentialregime. According to the DPI dataset, Moldova shifted from a presidential system to the assembly-electedform of government in 2001.

11In the regressions, we do not control for being a low income countries with the dummy variablebecause we dropp from the sample a significant part of them.

13

rule, no statistically significant evidence is found, confirming previous results. Estim-ates of the model with the interaction term show that the coefficients of PRPRES andMDMPRES remain negatively and significantly correlated with RegQual. These res-ults are robust because the F-statistic shows that the coefficients α, φ and δ are jointlydifferent from zero, validating the interaction model for these finding.

Analogous results are found when the threshold is raised to the 20th percentile,thus excluding other 6 countries with an average GDP per capita below 1,000 US$.12

Re-estimating model (1) with a sub-sample of 66 countries, the significant impact ofthe presidential regime disappears in most of the regressions. The interaction termPRPRES is again negatively and significantly correlated with RegQual. However, theF-statistic shows that the interaction model does not hold for these results.13

Re-running the panel regressions with a dummy that assumes value 1 if the countryis a member of the Organisation for Economic Cooperation and Development (OECD),and zero otherwise, our main results do not change. However, the dummy performspoorly because its coefficient is never significant.

The robustness check based on sub-samples of our dataset tends to reduce the sig-nificants of the control variables. This might depend on the reduced efficiency of theSys-GMM estimator when the number of observations becomes smaller.

Although the removal from the sample of the countries with the lowest levels ofthe GDP per capita makes the effects produced by the institutional combination lesspronounced, the direction of these effects remains the same, confirming our main results.

6 Concluding remarks

The ability of the government to devise and implement sound policies depends on thestructure of political institutions, such as the electoral rule and the political regime.Our analysis shows the importance of the electoral rules in determining the impact ofpolitical regimes on the quality of the government. We maintain that it is misleading toseparately evaluate the impact of electoral rules and political regimes, as it is from theirinteraction that the institutional conditions for the government to operate are shaped.Our empirical investigation shows that on average the presidential system tend to reducegovernment effectiveness and regulatory quality in democratic countries. However, thisform of government performs better when combined with a majoritarian electoral rulerather than a proportional rule.

Our work provide some evidence and guidance on the debate on constitutional re-forms, suggesting that given a political system the choice of the electoral rules wouldmake a difference for the ability of the government operate.

12We exclude Georgia, India, Mongolia, Papa New Guinea, Sri Lanka and Ukraine from the sample.A presidential regime is adopted by Georgia, Mongolia, Sri Lanka and Ukraine.

13The estimation results are available upon request.

14

Table 3: Estimation results of model (1)

GovEff(1) (2) (3) (4) (5) (6)

MAJ 0.160 0.121(1.38) (0.88)

PR -0.154 -0.027(-1.57) (-0.24)

MDM (log) -0.054 -0.013(-1.41) (-0.34)

PRES -0.197* -0.164 -0.216* -0.036 -0.080 -0.085(-1.72) (-1.63) (-1.73) (-0.35) (-0.76) (-0.78)

Party age 0.0004 0.0005 0.0002 -0.0001 0.0002 0.0001(0.62) (0.72) (0.24) (-0.10) (0.31) (0.19)

Pop (log) -0.004 -0.041 -0.021 0.040 0.026 0.028(-0.15) (-1.05) (-0.68) (1.24) (0.68) (0.87)

Urban Pop growth -0.074** -0.061** -0.065* -0.055 -0.039 -0.041(-2.38) (-2.11) (-1.95) (-1.41) (-1.16) (-1.14)

Religion homog. 0.003 0.003* 0.003* -0.0004 -0.0005 -0.0005(1.59) (1.67) (1.77) (-0.15) (-0.18) (-0.18)

Ethnic homog. -0.007* -0.006 -0.006*(-1.78) (-1.53) (-1.69)

GDP pc (log) 0.053 0.055 0.095 0.163 0.176* 0.180*(0.67) (0.72) (1.24) (1.57) (1.79) (1.69)

Openness -0.0002 -0.0005 -0.0002 0.002* 0.002 0.002(-0.20) (-0.59) (-0.23) (1.82) (1.43) (1.65)

CPI 0.001 0.001 0.001 0.002 0.001 0.002(1.05) (1.09) (0.97) (1.46) (1.39) (1.39)

Autonomy -0.044 -0.027 -0.058 -0.243 -0.212 -0.226(-0.38) (-0.29) (-0.53) (-1.33) (-1.21) (-1.21)

Demo 0.004 0.005 0.003 -0.001 -0.002 -0.002(1.40) (1.56) (1.20) (-0.32) (-0.40) (-0.53)

Low income 0.032 -0.018 -0.005 0.080 0.081 0.081(0.28) (-0.13) (-0.04) (0.50) (0.46) (0.49)

GovEfft−1 0.721*** 0.694*** 0.684*** 0.732*** 0.697*** 0.704***(9.52) (9.49) (8.83) (6.36) (5.31) (5.82)

F-test 0.000 0.000 0.000 0.000 0.000 0.000m1 test 0.000 0.000 0.000 0.000 0.000 0.000m2 test 0.165 0.194 0.176 0.186 0.181 0.180Hansen J test 0.125 0.119 0.127 0.135 0.130 0.130Diff-in-Hansen test 0.247 0.261 0.340 0.248 0.269 0.249Instrument No. 78 78 78 65 65 65Groups No. 80 80 80 67 67 67Obs. No. 1045 1045 1040 834 834 834

Note: Instruments for the first differences equation: GovEfft−2, GDP pct−2 (collapsed),Demot−2 (collapsed), Demot−3 (collapsed), Party aget−2, Opennesst−2 (collapsed) in (1)-(6);GDP pct−3 (collapsed), Party aget−3 in (1)-(3); Opennesst−3 (collapsed) in (4)-(6). Instru-ments for levels equation: ∆GovEfft−1, ∆GDPpct−1 (collapsed), ∆Demot−1 (collapsed),∆Partyaget−1, ∆Opennesst−1 (collapsed), constant term in (1)-(6). Standard errors robustto heteroschedasticity. t-statistics in parenthesis. p-value is reported for the tests. Significantat level ∗∗∗1%, ∗∗5%, ∗10%.

15

Table 4: Estimation results of model (1)

RegQual(1) (2) (3) (4) (5) (6)

MAJ -0.189 -0.087(-1.39) (-0.54)

PR -0.075 -0.075(-0.79) (-0.621)

MDM (log) 0.012 0.016(0.29) (0.400)

PRES -0.378** -0.254** -0.313** -0.292** -0.209* -0.268**(-2.59) (-2.16) (-2.41) (-2.05) (-1.95) (-2.29)

Party age 0.001 0.001 0.001 0.002 0.002 0.002(1.04) (1.22) (1.05) (1.41) (1.40) (1.31)

Pop (log) 0.021 0.014 0.031 0.038 0.029 0.039(0.51) (0.37) (0.82) (0.87) (0.69) (0.88)

Urban Pop growth -0.020 -0.071 -0.065 -0.095* -0.112* -0.115**(-0.45) (-1.62) (-1.39) (-1.73) (-1.96) (-2.02)

Religion homog. 0.001 0.002 0.001 0.001 0.001 0.000(0.28) (1.01) (0.65) (0.31) (0.71) (0.26)

Ethnic homog. -0.002 -0.001 -0.002(-0.44) (-0.34) (-0.56)

GDP pc (log) -0.095 -0.142 -0.126 -0.035 -0.061 -0.061(-1.12) (-1.60) (-1.38) (-0.34) (-0.60) (-0.57)

Openness 0.002 0.002* 0.002 -0.000 -0.000 -0.000(1.57) (1.71) (1.30) (-0.09) (-0.08) (-0.26)

CPI 0.002** 0.003** 0.002** 0.002 0.002* 0.002(2.32) (2.61) (2.30) (1.60) (1.78) (1.48)

Autonomy 0.086 -0.001 0.030 0.330* 0.247 0.290(0.86) (-0.01) (0.33) (2.17) (1.54) (1.50)

Demo 0.004 0.006* 0.0047 -0.0014 0.001 -0.001(1.21) (1.74) (1.40) (-0.34) (0.13) (-0.17)

Low income -0.184 -0.218 -0.181 -0.093 -0.128 -0.109(-1.03) (-1.24) (-1.02) (-0.59) (-0.79) (-0.68)

RegQualt−1 0.688*** 0.750*** 0.732*** 0.826*** 0.851*** 0.847***(7.89) (9.53) (9.61) (10.26) (10.07) (9.41)

F-test 0.000 0.000 0.000 0.000 0.0000 0.000m1 test 0.000 0.000 0.000 0.0000 0.000 0.000m2 test 0.848 0.605 0.683 0.404 0.371 0.337Hansen J test 0.180 0.202 0.173 0.141 0.189 0.172Diff-in-Hansen test 0.546 0.374 0.299 0.119 0.160 0.131Instrument No. 77 77 77 61 61 61Groups No. 80 80 80 67 67 67Obs. No. 1045 1045 1040 834 834 829

Note: Instruments for the first differences equation: RegQualt−2, Demot−2 (collapsed)in (1)-(6); RegQualt−3, GDP pct−3 (collapsed), Demot−3 (collapsed), Party aget−2,Opennesst−3 (collapsed) in (1)-(3); GDP pct−2 (collapsed), Party aget−3, Opennesst−2

(collapsed) in (4)-(6). Instruments for levels equation: ∆RegQualt−1, ∆Demot−1 (col-lapsed), constant term in (1)-(6); ∆Partyaget−1, ∆GDPpct−2 (collapsed), ∆Opennesst−2

(collapsed) in (1)-(3); ∆Partyaget−2, ∆GDPpct−1 (collapsed), ∆Opennesst−1 (collapsed)in (4)-(6); Standard errors robust to heteroschedasticity. t-statistics in parenthesis. p-valueis reported for the tests. Significant at level ∗∗∗1%, ∗∗5%, ∗10%.

16

Table 5: Estimation results of model (2)

GovEff(1) (2) (3) (4) (5) (6)

MAJ -0.066 -0.094(-0.36) (-0.47)

MAJPRES 0.950** 0.768*(2.03) (1.73)

PR -0.009 0.075(-0.07) (0.51)

PRPRES -0.405** -0.367(-2.36) (-1.58)

MDM -0.041 0.009(-0.89) (0.17)

MDMPRES -0.034 -0.059(-0.42) (-0.63)

PRES -0.315* 0.139 -0.158 -0.075 0.231 0.016(-1.79) (0.78) (-0.85) (-0.60) (1.07) (0.07)

Party age -0.0002 0.001 0.0001 -0.0001 0.001 0.0001(-0.22) (0.86) (0.16) (-0.15) (0.82) (0.15)

Pop (log) -0.032 -0.042 -0.022 0.011 0.009 0.023(-0.85) (-0.86) (-0.72) (0.28) (0.23) (0.75)

Urban Pop growth -0.086** -0.083** -0.060** -0.061 -0.055 -0.033(-2.11) (-2.30) (-1.96) (-1.23) (-1.17) (-0.89)

Religion homog. 0.005* 0.003 0.003* 0.0004 -0.001 -0.001(1.87) (1.13) (1.69) (0.14) (-0.24) (-0.27)

Ethnic homog. -0.007* -0.005 -0.006(-1.81) (-1.24) (-1.57)

GDP pc (log) 0.133 0.104 0.109 0.228** 0.180* 0.197*(1.13) (1.15) (1.37) (2.05) (1.74) (1.85)

Openness -0.001 -0.0005 -0.0003 0.001 0.002 0.001(-1.11) (-0.49) (-0.35) (0.98) (1.35) (1.29)

CPI 0.001 0.001 0.001 0.002* 0.002 0.002(1.00) (1.10) (0.93) (1.70) (1.61) (1.36)

Autonomy -0.033 -0.042 -0.066 -0.210 -0.251 -0.252(-0.22) (-0.36) (-0.67) (-0.97) (-1.32) (-1.30)

Demo 0.002 0.004 0.003 -0.002 -0.001 -0.002(0.51) (0.87) (1.06) (-0.51) (-0.30) (-0.56)

Low income 0.018 -0.029 0.013 0.065 0.001 0.092(0.13) (-0.17) (0.10) (0.37) (0.00) (0.52)

GovEfft−1 0.606*** 0.651*** 0.678*** 0.623*** 0.675*** 0.687***(5.00) (8.09) (8.64) (4.97) (5.45) (5.83)

α = φ = 0 0.077 0.043 0.364 0.187 0.267 0.773α = φ = δ = 0 0.078 0.048 0.235 0.295 0.410 0.738F-test 0.000 0.000 0.000 0.000 0.000 0.000m1 test 0.000 0.000 0.000 0.001 0.000 0.000m2 test 0.266 0.236 0.183 0.276 0.265 0.192Hansen J test 0.199 0.088 0.106 0.178 0.180 0.138Diff-in-Hansen test 0.318 0.145 0.299 0.355 0.306 0.289Instruments No. 78 78 78 65 65 65Groups No. 80 80 80 67 67 67Obs. No. 1045 1045 1040 834 834 829

Note: Instruments for the first differences equation: GovEfft−2, GDP pct−2 (collapsed),Demot−2 (collapsed), Demot−3 (collapsed), Opennesst−2 (collapsed) in (1)-(6); GDP pct−3

(collapsed), Party aget−3 in (1)-(3); Opennesst−3 (collapsed) in (4)-(6). Instruments forlevels equation: ∆GovEfft−1, ∆GDPpct−1 (collapsed), ∆Partyaget−1, ∆Demot−1 (col-lapsed), ∆Opennesst−1 (collapsed), constant term in (1)-(6). Standard errors robust toheteroschedasticity. t-statistics in parenthesis. p-value is reported for the tests. Significantat level ∗∗∗1%, ∗∗5%, ∗10%.

17

Table 6: Estimation results of model (2)

RegQual(1) (2) (3) (4) (5) (6)

MAJ -0.212* -0.040(-1.66) (-0.18)

MAJPRES 0.001 -0.135(0.00) (-0.25)

PR 0.117 0.015(0.93) (0.11)

PRPRES -0.437** -0.232(-2.27) (-1.06)

MDM 0.110* 0.040(1.88) (0.93)

MDMPRES -0.183* -0.065(-1.80) (-0.79)

PRES -0.389*** 0.001 -0.061 -0.274 -0.065 -0.171(-2.93) (0.00) (-0.40) (-1.57) (-0.43) (-1.02)

Party age 0.001 0.001* 0.001 0.002 0.002* 0.002(1.04) (1.67) (0.63) (1.39) (1.66) (1.32)

Pop (log) -0.003 -0.009 -0.003 0.046 0.021 0.036(-0.10) (-0.26) (-0.08) (0.90) (0.47) (0.73)

Urban Pop growth -0.006 -0.060 -0.028 -0.094* -0.122** -0.116**(-0.15) (-1.61) (-0.71) (-1.72) (-2.12) (-2.06)

Religion homog. -0.000 -0.001 -0.001 0.001 0.001 0.001(-0.11) (-0.29) (-0.26) (0.28) (0.44) (0.29)

Ethnic homog. -0.001 -0.001 -0.002(-0.28) (-0.29) (-0.55)

GDP pc (log) -0.087 -0.092 -0.050 -0.033 -0.049 -0.035(-1.13) (-1.03) (-0.56) (-0.32) (-0.48) (-0.31)

Openness 0.001 0.0004 -0.0001 0.0001 -0.001 -0.001(0.97) (0.41) (-0.09) (0.03) (-0.44) (-0.43)

CPI 0.002** 0.003*** 0.002** 0.002 0.002* 0.002(2.54) (2.64) (2.06) (1.51) (1.90) (1.48)

Autonomy 0.129 0.008 0.061 0.333** 0.196 0.275(1.37) (0.09) (0.56) (2.08) (1.26) (1.46)

Demo 0.004 0.004 0.004 -0.001 -0.0004 -0.001(1.28) (1.00) (1.04) (-0.34) (-0.09) (-0.28)

Low income -0.146 -0.141 -0.032 -0.081 -0.143 -0.079(-0.86) (-0.81) (-0.18) (-0.50) (-0.86) (-0.46)

RegQualt−1 0.705*** 0.766*** 0.681*** 0.831*** 0.849*** 0.824***(8.60) (9.49) (7.44) (10.25) (9.37) (9.28)

α = φ = 0 0.230 0.057 0.127 0.854 0.535 0.599α = φ = δ = 0 0.037 0.010 0.045 0.186 0.221 0.113F-test 0.000 0.000 0.000 0.000 0.000 0.000m1 test 0.000 0.000 0.000 0.000 0.000 0.000m2 test 0.960 0.858 0.882 0.413 0.362 0.334Hansen J test 0.154 0.142 0.081 0.127 0.365 0.237Diff-in-Hansen test 0.771 0.336 0.366 0.114 0.362 0.212Instruments No. 77 77 77 61 61 61Groups No. 80 80 80 67 67 67Obs. No. 1045 1045 1040 834 834 829

Note: Instruments for the first differences equation: RegQualt−2, GDP pct−2 (collapsed),Demot−2 (collapsed), Opennesst−2 (collapsed) in (1)-(6); Party aget−2, Opennesst−3 (col-lapsed) in (1)-(3); Party aget−3 in (4)-(6). Instruments for levels equation: ∆RegQualt−1,∆GDPpct−1 (collapsed), ∆Demot−1 (collapsed), ∆Opennesst−1 (collapsed), constant termin (1)-(6); ∆Partyaget−1 in (1)-(3); ∆Partyaget−2 in (4)-(6). Standard errors robust toheteroschedasticity. t-statistics in parenthesis. p-value is reported for the tests. Significantat level ∗∗∗1%, ∗∗5%, ∗10%.

18

Data source and detailed description of variables

Variable Data description Data source

GovEff Perceptions of the quality of publicservices, the quality of the civil ser-vice and the degree of its independ-ence from political pressures, thequality of policy formulation andimplementation, and the credibilityof the government’s commitment tosuch policies.

The Worldwide Governance Indic-ators (Kaufman et al., 2010), 2012.

RegQual Perceptions of the ability of thegovernment to formulate and imple-ment sound policies and regulationsthat permit and promote privatesector development.

The Worldwide Governance Indic-ators (Kaufman et al., 2010), 2012.

PR 1= if candidates are elected accord-ing on the percent of votes receivedby their party; 0 = otherwise.

The World Bank Development Re-search Group, Database of PoliticalInstitutions 2012 (DPI2012).

MAJ 1= if legislators are elected usinga winner-take-all/first past the postrule; 0= otherwise.

DPI2012.

MDM Mean district magnitude is theweighted average of the number ofrepresentatives elected by each con-stituency. When this information isnot available, in the DPI database,MDM corresponds to the numberof seats divided by the number ofconstituencies. For further detailssee the DPI codebook.

DPI2012.

PRES 1= Presidential system; 0=other-wise.

DPI2012.

Party age The average of the ages of the 1stgovernment party, 2nd governmentparty, and 1st opposition party, orthe subset of these for which age ofparty is known.

DPI2012.

19

Variable Data description Data source

Pop Population, total. The World Bank World Develop-ment Indicators (WDI).

Urban Pop growth Urban population growth (annual%).

The World Bank WDI.

Religion homog. Largest religious group (% of thepopulation).

The CIA World Fact Book.

Ethnic homog. Largest ethnic group (% of the pop-ulation).

The CIA World Fact Book.

GDP pc Gross domestic product per capita,computed at constant (2000) US$.

The World Bank WDI.

Openness Sum of import and export as per-centage of gross domestic product.

The World Bank WDI.

CPI Consumer price index (2005 = 100). The World Bank WDI.Autonomy 1= constitutionally autonomous or

self-governing region; 0= otherwise.DPI2012.

Demo The indicator corresponds to’Tensys’ variable; if the executiveindex of electoral competitiveness(EIEC) is below 6, the country isdeemed autocratic or a country inwhich democratic institutions arenot consolidated and leadership ispersonality-based. In this case, thesystem is as old as the executive?years in office. If EIEC is 6 or 7,then Tensys records how long thishas been the case.

DPI2012.

20

Appendix

Countries include in our sample

Albania, Argentina, Australia, Austria, Bangladesh, Barbados, Belgium, Belize, Benin, Boliva, Bot-swana, Brazil, Bulgaria, Canada, Chile, Colombia, Costa Rica, Croatia, Cyprus, Czech Republic, Den-mark, Dominican Republic, Ecuador, El Salvador, Finland, France, Georgia, Germany, Ghana, Greece,Grenada, Guyana, Honduras, Hungary, Iceland, India, Ireland, Israel, Italy, Jamaica, Japan, Latvia,Lithuania, Luxembourg, Macedonia FYR, Madagascar, Malawi, Mali, Malta, Mauritius, Moldova, Mon-golia, Mozambique, Namibia, Netherlands, New Zealand, Nicaragua, Norway, Panama, Papua NewGuinea, Paraguay, Poland, Portugal, Romania, Samoa, Slovak Republic, Slovenia, South Africa, Spain,Sri Lanka, Suriname, Sweden, Switzerland, Trinidad and Tobago, Turkey, Ukraine, United Kingdom,United States, Uruguay, Venezuela.

21

References

Aghion, P., A. Alesina, and F. Trebbi (2004). Endogenous political institutions. Quarterly Journal ofEconomics 119 (2), 565–611.

Arellano, M. and S. Bond (1991). Some tests of specification for panel data: Monte carlo evidence andan application to employment equations. The Review of Economic Studies 58 (2), 277–297.

Arellano, M. and O. Bover (1995). Another look at the instrumental variable estimation of error-components models. Journal of Econometrics 68 (1), 29–51.

Bartolini, D. (2014). Administrative fragmentation and performance of OECD TL2 regions. RegionalDevelopment Policy Working Paper Series, forthcoming.

Bartolini, D. and R. Santolini (2013). Fiscal autonomy and quality of governance in oecd countries.Economics Bulletin 33 (1), 706–713.

Beck, T., G. Clarke, A. Groff, P. Keefer, and P. Walsh (2001). New tools in comparative politicaleconomy. World Bank Economic Review 15 (1), 165 – 176.

Blundell, R. and S. Bond (1998). Initial conditions and moment restrictions in dynamic panel datamodels. Journal of Econometrics 87 (1), 115–143.

Bun, M. J. G. and F. Windmeijer (2010). The weak instrument problem of the system gmm estimatorin dynamic panel data models. Econometrics Journal 13 (1), 95–126.

de Mello, L. and M. Barenstein (2001). Fiscal decentralization and governance: a cross-country analysis.IMF Working Paper 01/71.

Duverger, M. (1954). Political parties. New York: Wiley.

Enikopolov, R. and E. Zhuravskaya (2007). Decentralization and political institutions. Journal of PublicEconomics 91, 2261 – 2290.

Kaufmann, D., A. Kraay, and M. Mastruzzi (2010, September). The worldwide governance indicators:Methodology and analytical issues. World Bank Policy Research Working Paper No. 5430.

Kunicova, J. and S. Rose-Ackerman (2005). Electoral rules and constitutional structures as constraintson corruption. British Journal of Political Science 35, 573–606.

Kyriacou, A. P. and O. Roca-Sagales (2011). Fiscal decentralization and government quality in theOECD. Economics Letters 111, 191 – 193.

Lardeyret, G. (1991). The problem with PR. Journal of Democracy 2 (3), 30–35.

Lijphart, A. (1991). Constitutional choices for new democracies. Journal of Democracy 2 (1), 72–84.

Lijphart, A. (1999). Government forms and performance in thirty-six countries. Yale University Press.

Linz, J. J. (1990). The perils of presidentialism. Journal of Democracy 1 (1), 51–69.

Mainwaring, S. (1993, 07). Presidentialism, multipartism, and democracy: The difficult combination.Comparative Political Studies 26 (2), 198–228.

Mainwaring, S. and M. Shugart (1997). Juan Linz, presidentialism, and democracy: a critical appraisal.Comparative Politics 29 (4), 449–471.

22

Panizza, U. (2001). Electoral rules, political systems, and institutional quality. Economics and Polit-ics 13 (3), 311–342.

Persson, T., G. Roland, and G. Tabellini (2004). How do electoral rules shape party structures, govern-ment coalitions, and economic policies? CESifo Working Paper No. 1115 .

Persson, T. and G. Tabellini (1999). The size and scope of government:: Comparative politics withrational politicians. European Economic Review 43 (4–6), 699–735.

Qian, Y. and B. R. Weingast (1997). Federalism as a commitment to preserving market incentives.Journal of Economic Perspectives 11 (4), 83–92.

Roodman, D. (2009). Practitioners’ corner. a note on the theme of too many instruments. OxfordBulletin of Economics and Statistics 71 (1), 135–158.

Samuels, D. J. and M. S. Shugart (2003). Presidentialism, elections and representation. Journal ofTheoretical Politics 15 (1), 33–60.

Shugart, M. S. and J. M. Carey (1992). Presidents and assemblies: Constitutional design and electoraldynamics. Cambridge University Press.

Treisman, D. (2002). Decentralization and the quality of government. UCLA working paper.

Weingast, B. R. (1995). The economic role of political institutions: market-preserving federalism andeconomic development. Journal of Law, Economics, and Organization 11 (1), 1–31.

23